Time-lapse microscopy study of noise in development Gritti, Nicola DOI 10.4233/uuid:23580f57-9ed1-4fa3-95ad-b8fe7cc8ba05 Publication date 2017 Document Version Final published version Citation (APA)
Gritti, N. (2017). Time-lapse microscopy study of noise in development. https://doi.org/10.4233/uuid:23580f57-9ed1-4fa3-95ad-b8fe7cc8ba05 Important note
To cite this publication, please use the final published version (if applicable). Please check the document version above.
Copyright
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons. Takedown policy
Please contact us and provide details if you believe this document breaches copyrights. We will remove access to the work immediately and investigate your claim.
This work is downloaded from Delft University of Technology.
Nicola Gri�
Time-lapse microscopy
study of noise
in development
se micr
osc
op
y s
tudy of noise in de
velopmen
t Nic
ola Gri�
STUDY OF NOISE IN
DEVELOPMENT
Nicola GRITTI
ISBN 978-94-92323-13-2
STUDY OF NOISE IN
DEVELOPMENT
PROEFSCHRIFT
ter verkrijging van de graad van doctor aan de Technische Universiteit Delft,
op gezag van de Rector Magnificus prof. ir. K.C.A.M. Luyben; voorzitter van het College voor Promoties,
in het openbaar te verdedigen op donderdag, 13 april 2017 om 12:30 uur
door
Nicola GRITTI
Laurea Magistrale in Fisica
Università degli Studi di Milano-Bicocca, Italië geboren te Ponte San Pietro, Italië
FOM Institute AMOLF Copromotor: Dr. J. S. van Zon FOM Institute AMOLF
Composition of the doctoral committee:
Rector Magnificus Chairperson Prof. dr. ir. S. J. Tans Promotor Dr. J. S. van Zon Copromotor Independent members:
Prof. dr. M. Dogterom Technische Universiteit Delft Prof. dr. S. J. L. van den Heuvel Utrecht University
Prof. dr. ir. E. J. G. Peterman Vrije Universiteit Amsterdam Dr. H. O. Youk Technische Universiteit Delft
Dr. M. Barkoulas Imperial College, London, United Kingdom Prof. dr. N. H. Dekker Technische Universiteit Delft,
reserve member
The work described in this thesis is part of the research program of the Stichting voor Fundamenteel Onderzoek der Materie (FOM)
which is financially supported by the
Nederlandse Organisatie voor Wetenschappelijke Onderzoek (NWO). This work was carried out at the
FOM Institute AMOLF Amsterdam
1 Introduction 7 1.1 The role of randomness in biology 8 1.2 Noise in developmental biology 14 1.3 The role of C. elegans in developmental biology 17 1.4 Time-lapse microscopy of C. elegans 19
1.5 Thesis outline 22
2 Experimental method and C. elegans development in microfabricated
chambers 25
2.1 Microfabrication of polyacrylamide hydrogel chambers and sample
preparation 27
2.2 Time-lapse microscopy setup 32 2.3 Larval development in microfabricated chambers 40
2.4 Conclusions 46
3 Lineaging of stem-cell-like divisions 49 3.1 Experimental design and data analysis 52 3.2 Timing of seam cell divisions 57 3.3 Mutant with variable seam cell lineage 58
3.4 Conclusions 61
4 Quantitative analysis of oscillatory gene expression 63 4.1 Experimental design and data analysis 66 4.2 Characterization of mlt-10 expression 68 4.3 Characterization of wrt-2 expression 71
4.4 Conclusions 76
5 Quantitative study of the dynamics of the AC/VU stochastic cell fate
decision 79
5.1 The AC/VU stochastic cell fate decision 81 5.2 Quantitative analysis of gene expression in fixed animals with smFISH 88
5.3 Gene expression dynamics by fluorescence time-lapse microscopy 92 5.4 Conclusions 105 Bibliography 109 Summary 123 Samenvatting 127 Acknowledgments 131
1
Introduction
The concept of randomness and chance has intrigued human beings since the oldest times. In ancient history, it was thought that events are affected by the choices of the gods, who then were responsible for the variability detected in nature. During the 3rd century BC, however, Greek philosophers argued that randomness has a more natural essence. Democritus, for instance, believed that it is closely related to ignorance. Any unexpected result has a plausible explanation, therefore randomness is due to the inability of humans to fully understand the nature of events [1]. On the other hand, Epicurus suggested that Nature itself is continuously affected by random events at the smallest atomic scales and is therefore intrinsically unpredictable [2].
Since these pioneering works, the concept of randomness has been forgotten for several centuries, in an era in which philosophers were mostly focused on finding a higher meaning to the human lives and struggles. However, by the end of the 18th century, many revolutionary discoveries led to the establishment of classical mechanics, which seemed to suggest that Nature is governed by deterministic laws. Therefore, it was thought that natural events can be predicted upon full knowledge of the initial conditions of the system.
It was only in the late 19thcentury that findings in electrodynamics and thermody-namics undermined this believe. Since then, many scientific fields, including but not limited to quantum mechanics, statistical physics and particle physics, developed that describe natural events as stochastic, probabilistic and affected by noise. The idea of Epicurus became again widely accepted: natural events are variable, not predictable and partially driven by random events. Since then, the concepts of randomness and noise have played a central role in many disciplines in science, such as mathematics, physics, chemistry, statistics and, last but not least, biology. In this thesis, we aim to contribute to the understanding of the origins, the roles and the effects of noise in biology.
1.1
The role of randomness in biology
Although there are several hypothesis on how life arose on our planet [3], it is widely accepted that the oldest form of life on Earth is at least 4 billion years old [4]. As Darwin hypothesized in 1859 [5], it has been recently found that all living organisms on Earth descend from a universal common ancestor that lived at least 3.5 billion years ago [6]. Such primitive forms of life were very different from the organisms that populate our planet nowadays, as they were often anaerobic, thus did not require presence of oxygen to survive. Instead, they relied on simple chemical reactions to perform nitrogen and carbon dioxide fixation to extract energy. Nevertheless, it has been recently established that part of the biological machinery that keeps all current living organisms alive is relatively similar to that of the primitive organisms populating Earth billion years ago [6].
Even if all living organisms share some traits with their primitive common ancestor, over the course of 4 billion years life has evolved in a wide variety of forms, from sub-micrometer sized bacteria to the honey fungus that stretches over 2.4 km in the Blue Mountains in Oregon. It is estimated that more than one trillion species live on Earth, and that we have only been able to describe 1.2 millions of them, which is a mere 0.0001% [7, 8].
It may not seem surprising that life on our planet shows such a huge variety. After all, over the last 4 billion years, several more or less catastrophic events have happened that forced living organisms to evolve and adapt in order to survive, sometimes causing massive extinctions. Natural selection has worked over billion years to privilege the fittest species and extinguish species that were not able to cope with the ever-changing environment and the predators surrounding them. However, it is surprising the extent at which each of these organisms have been able to optimize their biology to the ecosystem and make the most out of the resources that the surrounding provided them. Each of the living organisms populating our planet can be seen as an almost perfect machine that receives inputs from the surrounding environment and reacts accordingly through a complicated internal machinery. Ultimately, this is what makes bacteria move towards food and antelopes run away from lions.
During the last century, biologists have worked hard trying to elucidate the complex mechanisms that allow all living organisms to survive. Probably the major advances in this respect are the discovery and description of deoxyribuncleic acid (DNA) as carrier of all the genetic information [9, 10] and the definition of the central dogma of molecular biology [11]. DNA is a molecule organized in a double helix structure and contained in every single cell. Each helix is a sequence of single monomers that can occur in four chemically different nucleotides. The combination of such nucleotides defines the genetic information specific to a particular organism. DNA molecules can be contained freely inside the cell wall, in which case we talk about prokaryotic cells, or can be enclosed in a sub-cellular structure, the nucleus, in which case we talk about eukaryotic cells. In most eukaryotic cells, the genetic
information is spread over different substructures called chromosomes. All multi-cellular organisms consists of such eukaryotic cells.
Despite the differences in its large-scale organization, DNA always contains all the instructions needed to perform the complex biological functions necessary to survive. Such instructions are encoded on the DNA, in the form of genes. The question of how this genetic information is translated into molecules able to perform complex functions is explained by the central dogma of molecular biology. As an approximate and brief version of it, the central dogma of molecular biology states that a single cell is able to replicate its own DNA entirely, translate parts of it into messenger ribonucleic acid (mRNA), and transcribe mRNAs into proteins. While replication of DNA is necessary for reproduction of the cell, the final products of this complex machinery are proteins, highly specialized molecules that are able to perform different tasks. Often multiple proteins combine to create a macromolecule able to perform a complicated task. For example, ribonucleic acid polymerase (RNAp) can bind to a particular site on the DNA and transcribe the gene found downstream into an mRNA. An even more complex molecule is the ribosome, the machine necessary to translate mRNAs into proteins. Many more proteins exists and each of them has a specific functional role in the survival of a cell.
During their entire life, cells need to perform many different tasks, each of them requiring the production of the corresponding proteins. It is therefore evident that not all the proteins must be produced at equal levels. A complex mechanism is exploited to regulate the production of proteins. At the earliest stage, gene expression can be tuned in order to produce more or less mRNAs. For instance, some molecules, called transcription factors, bind on specific parts of DNA, called promoters, to stimulate or inhibit binding of RNAp and therefore regulate the expression of the gene downstream on the DNA. Secondary, translational regulation can be used to suppress production of proteins. For instance, single mRNA molecules can be actively degraded.
1.1.1
Cell biology and thermal fluctuations
Considering the discovery of the central dogma of molecular biology and the theory of natural selection, it is natural to assume that all observable variability among living organisms can be fully explained by genetic and/or environmental variability. However, a third element is formed by stochastic variability, which causes random variations even in two genetically identical individuals exposed to exactly the same environmental conditions. To understand the origin of this source of variability, we need to give a closer look to the building blocks of life. The typical dimension of a cell is on the order of a few micrometers, while proteins size is on the nanometers scale. At these scales, thermal fluctuations are omnipresent and have a major effect on the molecular dynamics. Individual molecules move inside the cell by Brownian motion, and the chemical reactions that trigger the basic biological processes happen only when two or more components come into contact. That is the reason why
biochemical processes are probabilistic. For instance, one can never predict when a single RNAp will bind to the promoter of a certain gene, but can only estimate a binding rate: the probability that this event will happen within a certain amount of time.
Thermal fluctuations are largely ignored when studying a macroscopic system because the number of molecules involved is so large that fluctuations are eventually averaged out. Therefore, a fully deterministic description of macroscopic systems is possible. However, certain molecules are present in the cell with as few as 10-100 copies. When dealing with such low numbers, molecular fluctuations, often referred to as noise, cannot be simply averaged out, and the slightest deviation in the motion of a single transcription factor can have a significant effect on the expression level of its target gene. It is therefore easy to imagine that molecular fluctuations have a major impact on many cellular processes such as gene expression, cell signaling and motility.
1.1.2
Gene expression noise
Historically, the first biological process in which stochasticity has been acknowledged is gene expression. Even though stochasticity in gene expression had been previously observed [12, 13], it was not until the late 1990s that scientists started to recognize that gene expression is indeed strongly stochastic [14]. After this preliminary work, many researchers started looking for more experimental evidence of such phenomenon.
In one of the first studies showing stochastic gene expression [16], a synthetic biology approach was used. The authors engineered a repressilator: a genetic circuit consisting of three different genes that was designed to generate gene expression oscillations. The network was designed in a circular fashion: each gene expresses a transcription factor able to repress the expression of the consecutive gene in the loop. The authors placed all genes in a plasmid, a small DNA molecule, and inserted it in E. colibacteria. The expected result of such a circular negative feedback loop is an oscillatory system, with expression of each gene cyclically activated over the course of several hours. By tagging one of the genes with the green fluorescent protein (GFP), the authors were able to follow the dynamics of one of the three genes over time and confirm the predicted oscillatory dynamics. Besides showing the power of synthetic biology, with which it is possible to study gene regulatory networks, they reported a noisy behavior of the fluctuations. In particular, they found considerable cell-cell variability both in the amplitude and the period of the oscillations. The authors suggested that such noisy behavior could possibly be explained by stochastic fluctuations of the components of the system.
Shortly after that, two milestone studies explored the causes of stochastic gene expression. In the first study, the concepts of intrinsic and extrinsic variability were formulated for the first time [15]. The authors inserted two copies of the same promoter in E. coli bacteria expressing two different fluorescent proteins (CFP and
A B
Figure 1.1: Gene expression fluctuations. (A) Expression of CFP (green) and YFP (red) in a population of genetically identical E. coli. (B) Schematic of the time series of CFP (green) and YFP (red) expression levels in a single cell in case of significant intrinsic noise. Figures were taken from [15].
YFP). Fluorescence intensity from bacteria belonging to the same colony was then recorded with dual color time-lapse fluorescence microscopy (Fig. 1.1). Following their description of intrinsic and extrinsic variability, extrinsic fluctuations due to the variability in the abundance of RNAp and ribosomes affect both promoters equally. On the other hand, intrinsic fluctuations due to the stochastic nature of gene expression affects each promoter independently. Therefore, if no intrinsic stochasticity is present, promoter activities in the same cell should perfectly correlate. However, the authors found that the amount of CFP and YFP in the same cell was highly variable, which then resulted in high cell-to-cell variability in the expression of both CFP and YFP (represented as cells with different red (YFP) and green (CFP) intensities in Fig. 1.1A). This shows that both sources of noise are significant, and that the impact of intrinsic variability on a single gene expression is as high as its extrinsic variability. In the second study, evidence arose that intrinsic variability is present also in eukaryotic cells [17], although it has a smaller effect. The authors suggested that this could be due to the larger amount of molecules present in eukaryotic cells, supporting the hypothesis that molecular fluctuations are more relevant when a low number of molecules is involved.
Following the work on gene expression noise, researchers started to intensively study the protein production process as a whole, starting from the expression of the gene to the translation of the mRNA. First, it was observed that the rate of transcription of a gene is not constant, but instead it is often activated in a transient way. This phenomenon is called transcriptional bursts, and has been observed both in bacterial and eukaryotic cells [18, 19]. Transcriptional bursts are likely due to a number of different factors. For instance, in eukaryotic cells, the structure of chromosomes is changing constantly, allowing transcription of a gene only when that part of the chromosome is in an open state. In prokaryotic cells, the sources of transcriptional bursts are most likely due to fluctuations in RNAp abundance and
error-correction mechanisms resulting in pauses in the synthesis of mRNAs [20, 21]. Similarly, translational bursts were observed [22]. In particular, it was shown that the translational rate is not constant over time, but instead it follows a series of stochastic bursts. This is due to the low number of mRNA molecules and to the fact that a large amount of proteins is produced by a single mRNA before it starts being degraded.
1.1.3
Noise in cellular decision-making
Despite the advances that such models and experiments represent in understanding the dynamics of gene expression, the question remains whether such stochasticity can impact the state and behavior of an entire cell.
As mentioned above, transcription factors regulate the expression of target genes. However, transcription factors themselves are proteins produced via the expression of other genes, which in turn are regulated by different transcription factors. As a consequence, the cell consists of a deeply interconnected gene regulatory network (GRN). Within the cell, multiple genes produce the right set of proteins necessary to process input signals and trigger cellular responses. When the network has a particular architecture, small changes in the input can result in massively different responses on the level of the entire cell. Such GRNs are called genetic switches.
A classical example of a genetic switch is the λ switch [23]. When the λ bacteriophage, a well-studied virus, infects an E. coli bacterium, the viral DNA is typically inserted into the host and starts being replicated together with the bacterial DNA without causing any damage. This state is called the lysogenic state. However, the system can also exist in a different state, called the lytic state. When the lytic state is triggered, the viral DNA is massively transcribed and hundreds of viruses are synthesized, causing the lysis of the host and allowing the viruses to escape and hunt for another host. The cellular state is light-sensitive: small doses of UV light cause transient upregulation of the expression of a gene, which in turns activates itself in a positive feedback loop and finally results in the induction of the lytic state. This was one of the first examples in which small changes in the input can cause a dramatic effect on the response of an entire cell.
In the previous example different cellular responses are triggered by subtle changes in the input of the GRN. However, intrinsically stochastic switches, i.e. cel-lular responses that are triggered randomly and are affected by stochastic fluctuations at the gene expression level, also play an important role in biology.
A well-known example of a stochastic switch is found in the soil bacterium Bacillus subtilis. When a population of genetically identical B. subtilis is exposed to unfavorable environmental conditions, such as starvation, each cell can assume a number of different fates. Some cells decide to lyse and release their genetic material, which can be used by the other cells as food source, therefore increasing the chances of population survival [25]. Other cells develop into spores, a dormant, non-growing and highly resistant state. As soon as environmental conditions improve, the
A B
Figure 1.2: Gene expression dynamics during the competence decision in B. subtilis. (A) Frames from a video of a competence event. A single cell that stochastically enters the competent state expresses CFP (red). Other cells express YFP (green). White cells developed into spores. (B) Time-series of both gene expression levels for a cell assuming competence fate (green and red lines) and a cell that did not (faint lines). Figures were taken from [24].
sporulation state is terminated and cells recover their normal growing behavior [26]. Another particular cell fate is competence, i.e. the ability to take up extracellular DNA released by lysing cells [24]. External DNA could be used as food source or might be integrated into the genome to try to adapt to the unfavorable conditions [27]. The mechanism driving the competence decision have been extensively studied. First it was shown that the decision to assume the competent fate is governed by a complex gene regulatory network exploiting positive and negative feedback loops, in which a major role is played by a gene called comK (Fig. 1.2A) [24, 28]. Next, it was shown that comK expression is intrinsically stochastic, and that such stochasticity drives the cell decision, in that large enough fluctuations are amplified by a positive feedback loop and result in the switch of the cell to the competent fate (Fig. 1.2B) [29]. Competence decisions are therefore intrinsically stochastic.
These results represent important evidence showing how noise at the gene expression level can greatly affect the state of an entire cell. Moreover, they show how noise can be beneficial for the survival of a population in an unpredictable environment.
These examples show that noise at the gene expression level can affect gene regulation in bacterial cells. These phenomena can have effects in decision-making processes, leading to stochastic responses of individual cells to the same environ-mental inputs. It is therefore natural to ask whether such stochasticity also affects the biology of multi-cellular organisms, particularly in the case of multi-cellular development. Intriguingly, significantly different phenotypic traits are commonly found between genetically identical multicellular organisms. For instance, even
identical human twins have different fingerprints and different risks of contracting diseases such as rheumatoid arthritis [30]. In the next section I will discuss recent studies on the role of noise in the biology of multi-cellular organisms.
1.2
Noise in developmental biology
Multi-cellular organisms development is a remarkably reliable and complex program during which, starting from a single-celled embryo, cell divisions and differentiation give rise to a fully developed adult capable of reproduction. During development, cells need to tightly control their positions and coordinate their behavior to be in the right place at the right time to perform their functions in the organism.
However, as we have discussed in the previous section, the most basic biological processes are intrinsically stochastic. All cells of a multi-cellular organism are likely subject to the same sources of stochasticity as for unicellular organisms. However, in contrast to unicellular organisms, in multi-cellular organisms a small number of cells can affect the behavior of several surrounding cells, for instance via cell-cell communication. For this reason, noise at the single cell level can have dramatic consequences for the entire organism.
The fact that such serious mistakes are extremely rare suggests that multi-cellular organisms have developed complex control mechanisms to reliably progress through development [31]. The ability of multi-cellular organisms to reliably develop despite noise is called robustness. The intrinsic conflict between stochasticity and developmental robustness raises the fundamental question of how noise is suppressed. In this section, I will discuss examples in which noise is detrimental and needs to be suppressed in order to ensure the correct development of the organism. Even though development is largely robust to noise, interestingly examples exist in which noise might not be detrimental but instead is thought to drive development. In this section, I will also discuss such examples. At the end, I will highlight the fundamental questions that need to be addressed in order to have a deep understanding of the role of stochasticity in developmental biology.
1.2.1
Robustness to developmental noise
Examples that have been extensively studied exist in which multi-cellular organisms have developed complex control mechanisms to suppress noise in order to reliably progress through development.
A classical example in which noise suppression mechanisms are exploited is the morphogenesis of the fruit fly Drosophila melanogaster embryos [34]. Morphogen-esis is the process by which an organized spatial distribution of cells is generated during embryonic development. During early embryogenesis of D. melanogaster, a morphogen protein called Bicoid is produced at the anterior pole, and diffuses into the embryo, thereby generating an exponentially decaying concentration gradient
A B
Figure 1.3: Stochasticity in developmental processes. (A) Expression of the final target gene responsible for intenstinal cell fate in wild-type (left) and mutant (right) animals. Figure taken from [32] (B) Cone cell in a human retina color-coded according to the photopigment chosen (red, green and blue). Figure taken from [33].
along the anteroposterior axis, with a characteristic length of about 100 µm. In order to generate the correct spatial pattern, single cells reliably detect their relative position along the gradient by measuring the local Bicoid concentration. Among other genes, Bicoid triggers the expression of hunchback, which then controls the expression of crucial downstream genes. The spatial profile of Hunchback (Hb) is strongly non-linear, with a steep drop in the middle of the embryo. Despite the noise in the Bicoid concentration, the authors found that the Hb profile had extremely low noise levels and that the position of the drop was remarkably precise. This suggests that stochastic cellular decisions due to intrinsic fluctuations are strongly suppressed. In particular, the authors suggested that neighboring cells are able to communicate in order to accurately estimate the Bicoid concentration.
The previous example suggests that gene expression noise and stochastic cell decisions are strongly suppressed during development of multi-cellular organisms. What happens when the fluctuations are not controlled and noise suppression fails? Already in 1925 researchers observed that some genetic mutations in the fruit fly Drosophila fimebris result in variable outcomes, with a fraction of individuals showing a wild-type phenotype, while the other part of the population shows a mutant phenotype [35, 36]. Recent studies examined the mechanistic origins of this phenomenon, called incomplete penetrance, in the intestinal cell fate specification of the nematode worm Caenorhabditis elegans [32]. The intestinal cell fate specification in C. elegans is regulated by a simple genetic circuit. By creating a mutant in which a key transcription factor was not expressed, the authors showed that the expression pattern of intermediate genes became highly variable and that the final target gene responsible for the cell fate specification assumed a bimodal distribution (Fig. 1.3A). These results show that the architecture of the GRN underlying intestinal induction is optimized to suppress noise and ensure proper cell specification.
These examples show that mechanisms exist to strongly suppress gene expression noise, leading to highly robust development of multi-cellular organisms.
1.2.2
Stochastic cell fate decisions
Although the robustness of developmental processes is often achieved by suppressing stochasticity, interesting examples exist that show how stochastic gene expression can actually be exploited to perform a specific developmental program. Often these mechanisms result in stochastic cell fate decisions, a process in which a cell differentiates in a random manner, by choosing one cell fate out of a repertoire of different fates [37].
One example of stochastic cell fate decisions is the photoreceptor selection in primates [33, 38]. Each of the 4 million cone cells in the human retina, for instance, chooses one type of photoreceptor out of three possible choices: red, green and blue, in what appears to be a random, cell-autonomous decision. The result is a random pattern of cell fates in the retina (Fig. 1.3B). Another example of cell-autonomous cell fate decision is olfactory receptor selection in mice [39]. In this much more complicated system, each olfactory neuron randomly expresses one gene out of ~1300 possible genes, exploiting a stochastic mechanism similar to the cone cells specification in the human retina.
Interestingly, some stochastic cell fate decisions also involve cell-cell commu-nication. In these cases, the stochastic process in one cell impacts that in the neighboring cells, and vice versa. For instance, in the fruit fly D. melanogaster, neuronal cells exploit signaling and feedback mechanisms to specify their fate [40]. This cell-cell interaction process results in a mutually exclusive and highly reliable cell fate assignment: when one particular cell randomly assumes a neuronal fate, all neighboring cells become epidermal cells.
These examples suggest that random fluctuations at the gene expression level are not detrimental, but instead can be exploited to drive development.
Taken together, these results suggest that multi-cellular organisms are subject to molecular fluctuations and that they have developed different mechanisms to reach robustness. In some cases, noise is efficiently suppressed, making the outcome of the developmental process almost deterministic. In other cases, organisms exploit noise to reach a variable but robust developmental outcome.
The intrinsic stochastic nature of the molecular players involved in the regula-tory network underlying developmental processes raises a number of fundamental questions:
• What are the sources of noise that impact development? How strong are their fluctuations?
• How can such fluctuations be suppressed in deterministic developmental processes to achieve a robust outcome?
• Do developmental processes exist that rely on and are driven by molecular noise? If so, how are molecular fluctuations amplified to impact the behavior of entire cells in the developing organism?
While a molecular biology approach will reveal the key components of the underlying regulatory network, the intrinsic fluctuations at the molecular level require a quantitative approach in order to address these questions. Moreover, to study a process as highly dynamic as development, an approach is needed to follow developmental processes over time. To this end, one needs to follow developing organisms with enough spatial and temporal resolution to detect the dynamics of the process at the single-cell level. However, most of the model systems for multi-cellular organisms, such as fruit flies, zebrafish and mice, have a large body size and a relatively slow development. Therefore, there are currently no techniques able to follow their development with enough spatial resolution and for more than few hours. In the next section, I will discuss an alternative model system that is extensively used for developmental studies: the nematode worm Caenorhabditis elegans. Moreover, I will argue that C. elegans represents an ideal model system to study the role of noise in development.
1.3
The role of C. elegans in developmental biology
In the early 1970s, with the pioneering work of Sydney Brenner, the nematode C. elegans has emerged as a model system in many fields in biology [41]. C. elegansis a soil nematode consisting of ~1000 cells. The full development from single-celled embryo to adult organism is ~48 hours long, and allows a 50 µm long egg to develop into a 1 mm adult organism. After 12 hours of embryonic development, a newly hatched larva grows for 36 hours into an adult organism. The post-embryonic development is divided in four larval stages (L1-L4), and at the end of each larval stage the animal enters a lethargus stage of 2 hours (Fig. 1.4). During this period, motility is strongly reduced and feeding stops. Eventually, a new cuticle is synthesized and the old one shed, an event called ecdysis, which then marks the start of the next larval stage.
Typically animals exist as hermaphrodite, which reproduce by self-fertilization. The progeny is therefore genetically identical to the mother except for rare random mutations. A single adult hermaphrodite can produce up to 350 offspring. At the same time, males are produced at low frequency (~0.1%), allowing for cross progeny. The simple genetics involved made C. elegans the first multi-cellular organism with a complete genome sequenced, revealing more than 19000 genes of which at least 40% code for proteins with homologous in higher organisms [42]. Many key regulatory genes in developmental and cell biology processes have so far been identified.
Thanks to its simple genetics, short life cycle, ease of maintenance and simple body plan, C. elegans is an ideal model system to perform developmental studies. In particular, all cells in the body can be imaged and identified using differential
L1 larva (~12 h, ~250 μm) L2 larva (~8 h, ~370 μm) L3 larva (~8 h, ~500 μm) L4 larva (~10 h, ~630 μm) embryo (~12 h, ~50 μm) adult (~2-3 weeks, ~1 mm)
Figure 1.4: Schematic of C. elegans life cycle at 22◦C. Embryos are laid approximately 3 hours after fertilization and continue developing for about 9 hours until hatching occurs. Numbers next to each animal indicate the length of each larval stage and the approximate length right after each ecdysis event. Figure adapted from Introduction to C. elegans anatomy chapter (WormAtlas).
interference contrast microscopy (DIC). As a result, all cell divisions have been detected and the full lineage from single-celled embryo to adult organism has been reconstructed [43].
Surprisingly, this study revealed that C. elegans development is largely invariant, in that cells divide and differentiate in a stereotypical manner. This suggests that C. elegansdevelopment, because of its extraordinary robustness, must be optimized to strongly suppress noise. For this reason, C. elegans is an ideal model system to study how deterministic developmental processes efficiently suppress molecular fluctuations. In this thesis, we did not directly examine mechanisms of noise suppression, but, as a starting point, we characterized the degree of variability in two developmental processes that show an invariant outcome: stem-cell like division patterns (Chapter 3) and gene expression oscillations during development (Chapter 4).
Even though C. elegans development is largely invariant, a few examples exist in which cells undergo stochastic cell fate decisions [44, 45]. The fact that these stochastic cell fate decisions take place within an environment inside the animal which is otherwise invariant, could potentially make it easier to pinpoint the sources of noise driving the cell fate decision. In Chapter 5 of this thesis, we perform a
quantitative analysis aimed to elucidate the sources of noise and the mechanisms underlying one of the best-understood stochastic cell fate decisions in C. elegans: the so-called AC/VU decision [37].
Because of the highly dynamic behavior and long duration of these processes, an approach able to follow single C. elegans larvae over the full post-embryonic development is an essential requirement. Moreover, because these developmental processes often involve a small number of cells, this approach should provide enough spatial resolution to follow single cells. In the next section, I will review a number of techniques currently used to perform time-lapse microscopy of C. elegans development and argue that none of them are suitable for the experiments we aim to perform.
1.4
Time-lapse microscopy of C. elegans
A first basic time-lapse microscopy protocol has been established in 1988 to follow C. eleganslarvae during development [46]. This technique is meant to aid the lineage analysis during the post-embryonic development of the animal [47], and requires manual loading of a single larva on a standard microscopy slide together with a small amount of E. coli bacteria as food source. Standard DIC microscopy techniques then allows researchers to image single nuclei and detect cell divisions.
Despite the technical simplicity of this technique, it is an inefficient way of imaging live animals, and presents several disadvantages. First, the whole process is manual, therefore extremely time-consuming and laborious. Second, only a single animal can be imaged at a time, thus severely limiting its throughput. This is in particular a problem when multiple animals in parallel need to be followed.
As an improved version of the previous technique, some paralysis-inducing drugs such as levamisole or sodium azide can be used to follow many animals in parallel by preventing them from moving [48]. However, such drugs also prevent the animal from feeding, leading to developmental arrest within a few hours. Therefore, a technique that combines the ability to follow many animals in parallel and to perform imaging over developmental time-scales is required. In order to tackle these challenges, several approaches have been recently developed that rely on microfluidic devices.
1.4.1
Microfluidic devices to study C. elegans
Microfluidics recently emerged as an important tool to perform microscopy analysis of C. elegans animals. In the last decade, microfluidic devices have been successfully used to study single bacterial cells [49], yeasts [50] and fruit fly embryos [51]. One of the great advantages of microfluidic devices is that environmental conditions can be controlled, possibly allowing for diffusion of chemicals at a specific time. Typically, a master mold with the desired pattern is created with soft-lithography
techniques. The mold is then used to create the inverse pattern in a soft material. PDMS (polydimethylsiloxane) is often the favorite material as it is transparent, permeable and biocompatible [52].
As many other biological fields, the C. elegans research field has been impacted by the advent of microfluidics, and a large variety of microfluidic devices has been recently designed [53, 54]. One typical application is worm handling, in which individual animals are loaded into narrow channels using an external flow [55]. Similarly, immobilization techniques can be used to perform high resolution imaging of sub-cellular structures. Immobilization can be achieved by flowing a cooling liquid in a separate channel [56], by deforming a flexible layer to compress the animal in a loading channel [57] or by reversibly gelating the surrounding fluid [58]. When immobilization is not necessary, animals can be placed in chambers or droplets, where food is provided and biological waste is removed through fluidic channels [59].
The majority of these techniques is geared towards the handling of adult animals, and they have been successfully used to perform behavioral studies, mutant screening and laser microsurgery [60–62]. However, as food is not provided to the animals, these techniques are not designed to sustain the full post-embryonic development. In the very few cases in which normal larval growth was supported, the techniques lacked the required spatial resolution to study sub-cellular processes [63, 64]. Moreover, the technological complexity of such techniques, consisting of multiple layers channels, liquid flow controls and surface treatments is a technological barrier for many C. elegans biology laboratories. In this thesis, I present an approach in which larvae are confined in small microfabricated chambers that have minimal impact on the animals in terms of mechanical stress (Chapter 2). At the same time, animals are able to freely move and feed in order to progress through the full post-embryonic development. Moreover, the minimal technological investments make our approach highly accessible to standard biology laboratories.
1.4.2
Microscopy techniques to image developmental processes
In addition to a device able to handle individual animals, a microscopy technique able to image developmental processes in live C. elegans larvae is needed. Development of multi-cellular organisms is driven by a variety of processes, such as tissue formation, cell division and gene expression. Thus, the main challenge in imaging developmental processes is to follow processes that occur simultaneously and at very different length scales. Specifically, this requires an imaging technique with a field of view large enough to image the whole animal, but still with enough spatial resolution to image sub-cellular events.
Many recent approaches try to match a large field of view with high spatial resolution. In fluorescence microscopy, axial resolution can be improved by optical
sectioning, a way to reject fluorescence light coming from out of focus focal planes [65]. Such techniques include confocal microscopy and two photon microscopy. Despite the high spatial resolution, these techniques require scanning of the incident beam over the area to be imaged. Therefore, there is a trade-off between field of view and acquisition speed. Often, high frame rate is achieved by illuminating only a small area, therefore limiting the field of view.
Techniques that are capable of large field of view imaging and high spatial resolution are limited. Recently, light sheet fluorescence microscopy (LSFM) proved to be an important technique capable of imaging large samples with high frame rate and medium resolution [66, 67]. With LSFM two objectives are used. An illumination objective shapes the laser beam in a thin sheet. A detection objective oriented in the orthogonal direction is used to collect the fluorescence light emitted from the sample. The main advantage of this technique is that phototoxicity, i.e. the toxicity damage caused in the sample by illuminating with high intensity light, is reduced compared to standard confocal techniques. That is because a single plane is illuminated at the time. Moreover, imaging can be fast, as no laser scanning is required, as in confocal microscopy. Light sheet fluorescence microscopy has been applied successfully to the study of embryonic development of Drosophila, Zebrafish and C. elegans [68, 69]. However, these techniques are not suited for imaging of C. elegans larvae, as they require a peculiar sample loading, in which the organism is embedded in a cylinder made of agarose gel and placed vertically in the microscope [70]. Few alternative configurations exist in which sample loading is more conventional, but these come at the cost of increased complexity in the design and have never been tested for developmental studies [71–73].
A novel and recent development of optical sectioning techniques is multifocal temporal focusing [74, 75]. This technique has the typical diffraction limited resolution of a two-photon imaging system and is able to image large field of views at high speed. The microscope design is completely equivalent to a standard confocal microscope, so that the sample can be loaded on the stage with a standard microscope slide. Temporal focusing has been used to perform whole brain calcium imaging at high spatial and temporal resolution in C. elegans and mice [76, 77], and it represents the most promising technique for fast, large field of view, high resolution imaging of freely moving animals. However, this technique is geared towards the imaging of thick opaque samples such as brain tissues. Therefore, when studying a simple transparent organism such as C. elegans, this technique is unnecessarily complicated. In fact, while optical sectioning is essential for thick samples imaging, it is often unnecessary for C. elegans imaging [78]. Instead, in order to image freely moving animals, we chose to develop a simpler technique that is optimized for acquisition of large field of views at high speed with high enough spatial resolution to follow single cells.
In this thesis, I present a technique to perform long-term fluorescence time-lapse microscopy of live C. elegans (Chapter 2). Our approach uses an imaging system
that is capable of acquiring large field of views at high frame rate but, at the same time, that is able to resolve single cells and sub-cellular structures. We use bright epi-fluorescence laser illumination to provide sharp and highly resolved fluorescence images with high temporal resolution in freely moving animals.
1.5
Thesis outline
In Chapter 2, I describe our fluorescence time-lapse microscopy technique: a combination of microfabricated chambers, wide-field fluorescence microscopy and image analysis. I provide the protocols used to perform the microfabrication, a detail description of our imaging system and a characterization of the performance of the setup in terms of single cell localization and fluorescence intensity quantification. Next, I test whether C. elegans larvae develop normally in our microfabricated chambers. In particular, I elaborate on the effects of microchamber dimensions and food availability on the development of C. elegans larvae. I find that individual animals develop normally in our microchambers as long as food is available.
In Chapter 3, I test the capability of our approach to follow single cells, by performing a quantitative lineaging study of cell divisions during development. In particular, I perform lineage analysis of seam cells, a model system for stem cell-like behavior, in multiple animals over the full post-embryonic development. To this end, I use fluorescence time-lapse microscopy of animals in which seam cell nuclei are fluorescently labeled. First, I perform a quantitative analysis of the time of division of all the seam cells in multiple wild-type animals. I find that some seam cells divide on average before others, suggesting that stage- and lineage-dependent temporal cues are responsible for the temporal regulation of seam cell divisions. Next, I characterize the variability in timing of divisions in this otherwise invariant developmental process. Moreover, by repeating the lineage analysis in mutant animals in which these cells do not follow the stereotypical division pattern, I show that stage- and lineage-dependent mechanisms are responsible for the correct execution of the stem-cell like divisions.
In Chapter 4, I prove that our setup is capable of quantifying gene expression levels, even in single cells, using fluorescence transcriptional reporter strains. To do so, I quantify the dynamics of expression of two genes which show an oscillatory behavior over the course of development. The first gene is expressed in the whole body of the animal, while the second is exclusively expressed in a number of cell nuclei. I find that expression of these two genes peaks once every larval stage. Next, I characterize the noise levels in the dynamics of these oscillations, and find that the times of the oscillation peaks show significant animal-to-animal variability. However, these times strongly correlate with the times of the closest molt, suggesting that a noise generated by a common source equally impacts the times of oscillations peaks and the times of ecdysis.
In Chapter 5, I show that our approach contributes to the understanding of a simple stochastic cell fate decision, the AC/VU decision, which relies on the
communication between two cells, referred to asαcells. Thanks to the well-known underlying gene regulatory network, we can use our time-lapse microscopy technique to study, for the first time, the dynamics of expression of the key molecular players involved in the decision. To this end, I use our fluorescence time-lapse microscopy technique on transcriptional reporter strains. Here, I aim to elucidate the sources of noise responsible for the AC/VU cell fate decision process. I show that, as previously reported, the birth order of theαcells biases the outcome of the process. However, I also show that other sources of noise must be responsible for the cell fate determination when the twoαcells are born at similar times. Next, I explore whether the stochastic expression of lag-2, one of the key components of the underlying gene regulatory network, before the time of births of theαcells, i.e. in their mother cells, could form this additional source of noise and, hence, bias the decision when theα cells are born at similar times. However, our results are not conclusive, leaving the identification of additional sources of noise an open question. At the end, I comment on future directions to address this open question.
2
Experimental method and C. elegans
development in microfabricated chambers
This chapter is part of the following publication: "N. Gritti, S. Kienle, O. Filina and J. S. van Zon, Long-term time-lapse microscopy of C. elegans post-embryonic development. Nat. Commun. 7:12500 doi: 10.1038/ncomms12500 (2016)."
Thanks to its simple body plan, short life cycle and transparency, C. elegans is an ideal model system to perform quantitative studies of developmental processes. Despite these advantages, there is currently no technique available to study the full post-embryonic development of individual living C. elegans with high spatial and temporal resolution. This limitation is due to the high motility of C. elegans larvae and their need to feed in order to properly develop. The standard time-lapse microscopy technique consists on imaging freely moving animals on a nematode growth medium (NGM) agar plate or on a microscope slide. Although this technique has been successfully used to perform neuronal and optogenetic studies [79, 80], it does not allow to image multiple animals in parallel, a key requirement when studying stochastic processes.
Current microfluidic devices designed to perform time-lapse microscopy of C. elegans are optimized to temporarily immobilize the animal, allowing high resolution imaging. Immobilization is accomplished using various strategies, in-cluding clamping [81, 82], compression [83, 84], cooling [85], nanometer size
beads [86] and gelation of the environment [87]. However, approaches based on microfluidics to immobilize animals have two major disadvantages. First, they are not designed to sustain imaging over developmental timescales. Second, their complicated designs often have undesired impacts on the animals during imaging. For instance, immobilization techniques have the potential to damage the cuticle of the larvae or to cause stress responses. In fact, previous works confirmed that physiological changes occur as a response to chemicals, mechanical stimulations and temperature changes [88–91].
In order to perform time-lapse microscopy over developmental timescales, single animals should be kept confined in an area that contains enough food to ensure proper development. Recently, solutions have been proposed that sustain larval development. Such systems use wells filled with agar gel and nutrients [63], growth chambers with inlet and outlet channels [64] and water-in-oil droplets filled with an aqueous solution containing nutrients [92]. This techniques have been successfully used to study motility of C. elegans larvae when exposed to different chemicals [93] and to quantify some markers of developmental progression such as growth rate and timing of larval stage transitions [64]. However, due to the large size of the compartments, these techniques are not compatible with high resolution imaging. Therefore, they do not provide sufficient spatial resolution to study developmental processes at the single-cell level.
We aimed to design a technique in which the impact of the device and imaging system on the development of the larvae is minimal. Ideally, this technique should be sufficiently simple as to be used by C. elegans biology laboratories, while at the same time allowing to perform time-lapse microscopy of post-embryonic C. elegans development with high spatial and temporal resolution over multiple animals in a parallel fashion.
Inspired by previous works [94], we explored the possibility of confining individual larvae in microfabricated compartments. First, we confine individual larvae in hydrogel-based microcompartments filled with E. coli bacteria as food source. The microfabricated chambers are large enough to provide sufficient food to sustain development for the full duration of the experiment, while small enough to fit in the field of view of the camera chip when using high magnification objectives to capture single-cell processes. Second, we use an imaging acquisition setup capable of acquiring images of larvae as they move inside the chamber. A combination of fast camera and bright illumination allows us to acquire sharp transmission and fluorescence images at different focal planes for each chamber even when animals are highly motile. Third, we use image analysis to reconstruct the dynamics of developmental processes between images at different time-points, such as cell divisions. This is a crucial part of our technique: instead of mechanically modifying the body shape and constraining the animal in a narrow channel, we image freely moving animals and exploit image processing techniques to obtain results that are independent of the body shape of the animal. For instance, we computationally straighten the animal body to define a convenient anteroposterior reference system.
This aids the analysis of the developmental processes we are interested in, including cell division detection and gene expression quantification, both in the whole animal and in single cells. One great advantage of our technique is the ease of use compared to previous microfluidics approaches: no multiple layers channels or liquid flows are required. Even though the idea of using microcompartments to confine individual animals is not new [94], it has only been used to study behavior over a limited time period and not to follow cellular dynamics over the full duration of development. Moreover, for reasons to be discussed in the next section, we propose to use an alternative material, polyacrylamide hydrogel, to the one previously used, agarose hydrogel.
In this chapter, I describe in detail our technique and show that it can sustain normal development. In Section 2.1, I describe how to fabricate microchambers in polyacrylamide hydrogel and how to confine C. elegans in such chambers. Next, I introduce the fast acquisition setup to perform volumetric imaging at ~100 fps in an automated fashion (Section 2.2). In the same section, I characterize the ability of our setup to quantify fluorescence signals and I briefly comment on the amount of data generated by our technique. Finally, in Section 2.3, I test whether wild-type C. eleganslarvae develop normally when confined in microchambers by quantifying several markers of developmental progression.
This chapter shows that we have developed a powerful new technique to perform time-lapse microscopy of freely moving and feeding C. elegans larvae in a parallel fashion and with high spatial and temporal resolution. In addition, our findings show that microchambers are able to sustain the full post-embryonic development of C. eleganslarvae.
2.1
Microfabrication of polyacrylamide hydrogel
chambers and sample preparation
Due to their tunable mechanical properties [95], hydrogels are the most common choice for the creation of microenvironments to study micro-organisms. In particular, agarose hydrogel has been succesfully used to confine bacteria for single cell studies [96–98]. The great advantage of this hydrogel is its permeability to chemicals, which provides control and uniformity of the microenvironment without the need of complicated flow systems. Recent experiments showed that agarose hydrogel microcompartments can be created to confine live nematode larvae for behavioral studies [94]. However, agarose is fragile and difficult to handle, especially in the thin layers required to create the microcompartments. Moreover, agarose consists mainly of water and galactase, which is easily metabolized by micro-organisms [99]. Therefore, the microenvironment to which single animals are exposed is subject to degradation.
has many practical advantages that make it an ideal material to create microenvi-ronments to confine biological samples. Polyacrylamide is widely used for DNA and protein electrophoresis, and thus it is easily accessible to biology laboratories. Polyacrylamide gels are ideal for microfabrication because they have highly tunable elastic properties and their micropatterning is easily accomplished by cost-effective techniques [100–102]. Thanks to its versatile mechanical properties, polyacrylamide is less brittle and easier to handle (fracture energy G ~10 - 50J ·m−2) than agarose (G
~0.1 - 6 J · m−2) [103]. Moreover, in contrast to agarose, the polymers that compose the gel are not metabolized, making it an ideal material to study biological samples in a stable microenvironment. Despite the fact that residual monomers in the gel are toxic, polyacrylamide is fully biocompatible when proper washing steps are used to remove the monomers [100]. Polyacrylamide hydrogel has been successfully used to fabricate highly controllable microenvironments to observe bacteria, yeast and, for short time periods, C. elegans [104].
Considering its ease of use and tunable mechanical properties, we chose poly-acrylamide hydrogel to create microfabricated chambers to confine larvae in a small area. Chambers are filled with food to sustain development over the course of the experiment. Therefore, chamber dimensions should be chosen based on the duration of the biological process of interest. The shape of the chambers should also be optimized: considering that most cameras have a squared chip, we designed squared microchambers to maximize the relevant area that is imaged. The depth of the chambers is also optimized. On the one hand, they are shallow enough to prevent C. eleganslarvae from moving in the axial direction, therefore minimizing the number of focal planes to be imaged. On the other hand, chambers are deep enough to prevent larger animals from being damaged by mechanical compression. To this end, the tunable mechanical properties of the hydrogel are beneficial, as we can optimize polyacrylamide stiffness such that it deforms to accomodate larger animals with modest compression in the axial direction.
2.1.1
Microfabrication
In our approach, we created a master mold with standard soft-litography techniques [105]. To pattern a 4 inch silicon wafer, the following protocol was used:
• The 4 inch silicon wafer was cleaned with isopropanol and cleared of dust particles before heating it on a hotplate at 150◦C and cooling it with nitrogen air (N2). Next, the silicon wafer was spin-coated with an epoxy resin (SU-8,
MicroChem, Fig. 2.1A). The viscosity of the epoxy resin was chosen and the speed of the coating was tuned to obtain homogeneous layers of 10 or 20 µm (Table 2.1).
• The silicon wafer was placed on a hotplate at 65◦C for 3 minutes and trans-ferred to a 95◦C hotplate for 5 minutes. The silicon wafer was then allowed to cool down at room temperature for about 30 minutes. This process, called soft
A
E
Top view Side view B Top view Side view
C Top view Side view D Top view Side view
2 hours silicon wafer SU8 layer foil mask UV light developer cuts mechanical clamps glass slide glass spacer hydrogel 50 ml falcon tube water F
Figure 2.1: Fabrication of microchambers. (A) A thin layer (10-20 µm) is spin coated on a 4 inch silicon wafer. The maximum speed is reached in two steps: first, to homogeneously spread the SU-8, 500 rpm speed is reached with an acceleration of 100 rpm/s, and it is maintained for 15 seconds. Then, the desired speed is reached with an acceleration of 500 rpm/s and maintained for 45 seconds (Table 2.1). (B) The silicon wafer is exposed to UV light. A foil mask is used to pattern the SU8. Exposure time used was 20 s. (C) The silicon wafer is immersed in SU-8 developer (Mr. Dev 600) for approximately 3 minutes to remove the SU-8 in excess. (D) The patterned silicon wafer is cut in 3 molds with the same size as a glass slide (26 mm x 76 mm). (E) Preparation of the polyacrylamide chambers. One mold is glued to a glass spacer with high vacuum grease. A 29:1 solution of polyacrylamide-bisacrylamide is poured in the cavity. A silanized glass slide is lowered on the solution and mechanically clamped. (F) After approximately 2 hours, the glass slide is removed. Each array of chambers is cut and washed in distilled water.
bake step, was needed to improve the adherence of the SU-8 to the surface of the wafer.
• Epoxy resins are negative photoresists, i.e. light-sensitive materials that cross-link when exposed to ultraviolet light (λ ≤ 400nm). We then used a foil mask and exposed the image to UV light, such that only part of the SU-8 layer cross-linked (Fig. 2.1B). The time the mask is exposed depends on the power of the illumination source, and in our case we used a 20 second exposure time at 25
Thickness (µm) Speed (rpm) SU-8 3010 SU-8 3025
3000 12 26
3500 10 22.5 4000 9.5 19.5
Table 2.1: Thickness of SU8 photoresist. The thickness of the photoresist depends on the maximum speed and on the viscosity of the SU-8 used. SU-8 3025 is more viscous than SU-8 3010. Thickness is in µm.
mW. The foil masks contained the desired pattern, in our case consisting of 9 arrays of 10x10 squared chambers. Final dimensions of the microchambers are equal to the dimensions of the structures on the mask.
• A post bake step was used to help adhesion of the cross-linked SU-8 to the silicon surface. The silicon wafer was treated at 65◦C for 1 minute and at 95◦C for 6 minutes. Subsequently, it was let to cool down at room temperature. • To remove the SU-8 in excess, we immersed the full silicon wafer in a chemical
solvent (Mr. Dev 600) for about 3 minutes, a process called development (Fig. 2.1C). The wafer was immersed for 10 seconds in another beaker with fresh developer and for another 10 seconds in isopropanol to stop the development. The patterned wafer was then dried with N2air.
• Once the silicon wafer was dry, it was hard baked at high temperature (200◦C) for 30 minutes on a hot plate to further cross-link the SU-8. This step also greatly improves the hardness of the micropattern, therefore preventing usage damage.
• The silicon wafer was then cropped with a diamond cutter to create the three final master molds, each containing 3 arrays of 10x10 structures (Fig. 2.1D). Once the patterned silicon wafer is ready, the fabricated mold can be used many times to create microfabricated chambers in polyacrylamide hydrogel. To prepare the polyacrylamide, we used a 29:1 ratio of acrylamide/bis-acrylamide solution (Bio-Rad) diluted to a final 10% concentration. Ammonium persulfate (Sigma, 0.1% of the volume) and TEMED (Sigma, 0.01% of the volume) were added to trigger the polymerization. The solution was poured in a cavity created by a hollowed standard microscope slide glued to the micropatterned silicon wafer with high vacuum grease. The polymerization reaction starts immediately upon addition of TEMED, therefore the solution must be poured in the cavity within 3-5 minutes. The cavity was then closed with a silanized glass slide and sealed by mechanical clamping (Fig. 2.1E). The solution was left to polymerize for at least 2 hours. When the gel was ready, the three arrays were cut with a scalpel from each cavity (Fig. 2.1F). After
polymerization, acrylamide monomers might still be present in the resulting gel. Acrylamide monomers are known to be a powerful neurotoxin, thus with the potential to negatively impact development of C. elegans larvae. To remove the monomers, at least 3 washing steps in distilled water of at least 3 hours each were necessary. When using fewer or shorter washing steps, we found that C. elegans larvae development was arrested during the first or second larval stage. Polyacrylamide gels could be stored in distilled water for at least 15 days without any visible degradation.
A
Top view
Side view
B
glass spacer
glass slide hydrogel
#1 coverslip animal bacteria lawn 10-25 μm 190-290 μm animal polyacrylamide hydrogel C objective
Figure 2.2: Sample preparation. (A) Schematic of the sample. A glass spacer is glued to a glass slide. The microchamber array is placed in the center of the cavity and single animals are transferred together with bacteria in the microchambers. The sample is closed with a coverslip ~100 µm thick. (B) The sample is mechanically clamped and sealed in a custom fabricated sample holder to prevent liquid evaporation. (C) The sample is placed upside down on the microscope to perform epi-fluorescence microscopy.
2.1.2
Sample preparation
To prepare the sample, we first transferred a microchamber array in M9 buffer for about 4 hours. C. elegans larvae were synchronized as follow: 15-20 adults were transferred on a fresh NGM agar plate spotted with E. coli bacteria (OP50) and allowed to lay eggs. After about 2 hours, the adults were transferred back to the original plate, so that the eggs left on the fresh plate were synchronized within a 2 hours period. The eggs on the fresh plate are then ready to be immediately transferred into the microfabricated chambers. A glass spacer with the same height as the polyacrylamide gel was glued to a glass slide using high vacuum grease (Fig. 2.2A). A single microchamber array was positioned on the glass slide, with the microchambers facing up. Excess liquid was removed with a tissue. The required time to transfer around 25-30 embryos is about 15 minutes. This is already long enough for the liquid to evaporate, causing the polyacrylamide gel to bend. To prevent this, a ~40 µl drop of M9 buffer was placed on the side and on the surface of the microchamber array, taking care to not let the liquid fill the chambers. Under a dissection microscope, a drop of bacterial suspension containing a single embryo was collected with an eyelash and transferred from the NGM agar plate into a single microchamber. To facilitate the release of the bacteria and embryo into the chamber, the eyelash was briefly dipped into the M9 drop prior to touching the microchamber. Ideally, the microchamber was already filled with enough OP50 bacteria after this step. However, if necessary, more bacteria were transferred to fill it completely. Subsequently, excess liquid was removed with tissue paper and the sample was closed with a #1 coverslip. The coverslip was lowered slow enough to avoid the formation of large air bubbles in between the polyacrylamide and the coverslip. The sample was then placed on a holder fabricated by the AMOLF mechanical workshop. The holder is optimized to minimize weight, thus allowing for rapid sample scanning along the axial direction. Moreover, the holder contains mechanical clamps to prevent liquid evaporation from the sample during the full duration of the experiment (Fig. 2.2B, C). Our design allowed us to load up to 50 chambers in a single sample.
2.2
Time-lapse microscopy setup
In order to study developmental processes with single cell resolution, we need to image individual chambers with high magnification and high numerical aperture (N.A.) objectives. However, C. elegans larvae can be highly motile. Therefore, in order to avoid motion blur due to the animal movement during a single image acquisition, we used an imaging system that provides bright illumination to reduce exposure time as much as possible. Moreover, to minimize movement between images acquired at different Z positions, we optimized our imaging system to quickly scan the sample along the Z direction.
hours of the post-embryonic development. Therefore, images typically need to be acquired on a timescale of minutes in order to resolve the dynamics of the developmental process. However, fluorescence microscopy causes phototoxicity in biological samples, especially when bright illumination (e.g. lasers) is used with long exposure times. A trade-off between exposure time, illumination power and time interval had to be determined for every experiment, but we generally found that 5-20 minutes time resolution and 1-10 ms exposure time do not lead to detectable phototoxicity. Moreover, as we want to perform parallel imaging of multiple animals, an automated system able to move through different chambers and acquire volumetric images is necessary.
2.2.1
Imaging setup
Instead of designing a completely new imaging system, we decided to optimize a commercially available inverted wide field microscope (Nikon Ti-E) to our needs (Fig.2.3A). We opted for a standard epi-fluorescence microscope with limited optical sectioning capability in favor of bright illumination and fast volumetric imaging. For all the experiments performed, the optical sectioning of our imaging setup was enough to resolve single cells and sub-cellular features. In fact, while optical sectioning techniques are essential for thick samples like vertebrates, they are often unnecessary for C. elegans imaging [78].
In our setup we used high magnification objectives (40X and 60X) and a camera with the largest possible chip (2048 x 2048 pixels Hamamatsu sCMOS Orca v2). The high numerical aperture of both objectives (N.A. = 1.3 for 40X, 1.4 for 60X) provides high spatial resolution, allowing to resolve sub-cellular structures (<1 µm). At the same time, the field of view of the camera was large enough to accommodate an entire chamber. In order to minimize the amount of UV light in the sample, which causes C. elegans larvae to move faster [106], transmission imaging was performed using a red LED (CoolLED p-100 615 nm).
In addition, we equipped our setup with two lasers for excitation of green (Coherent OBIS LS 488-100) and red fluorophores (Coherent OBIS LS 561-100). In contrast to standard lamp illumination systems, lasers have a much narrower bandwidth and a much higher intensity (80-100 mW). With these lasers, even very short exposure times (1-10 ms) were enough to have a high signal-to-noise ratio for all the strains studied. Moreover, as C. elegans larvae are highly motile even in absence of UV light (peaking at 50 µm s−1 for reversals), short exposure times minimize or even eliminate motion blur of fluorescently-labeled cells during acquisition. The two laser beams were combined in a single optical path with a dichroic mirror (Semrock LM01-503-25). In order to illuminate the microfabricated chambers as homogeneously as possible, we expanded the original laser beam from 0.7 mm to 36 mm with a telescope composed of two achromatic lenses of 10 and 500 mm focal length (Thorlabs), respectively. The expanded beam was then aligned through dielectric mirrors (Thorlabs) to enter the back aperture of our Nikon Ti-E inverted
